In the European internal market, the transport of people, goods and raw materials takes place largely by road. In the so-called D-A-CH countries Germany, Austria and Switzerland, the ageing transport network is heavily affected by a rapidly growing traffic volume. The frequency and density of inspections is increasing - resulting in intensive maintenance measures, ideally systematically planned in pavement management systems.
The condition recording of a road forms the basis for the planning of maintenance measures. Special measurement vehicles - equipped with cameras, but also increasingly with laser scanners - record the distress of road surfaces as part of measurement campaigns. While the acquisition is fast, the manual evaluation of the acquired data is proving to be more and more a bottleneck in the process chain.
3D-AI: Fraunhofer IPM uses KI for efficient data evaluation
The Deep Learning Framework 3D-AI developed by Fraunhofer IPM evaluates structural characteristics in measurement data faster, more efficiently and more objectively than it is possible with previous methods. The automated detection of structural features in the measurement data uses modern methods of machine learning, based on an artificial neural networks (ANN). Complex learning algorithms based on the concept of »deep learning« with ANN are used to evaluate the data. These algorithms are superior to the traditional methods of object recognition.
In addition to the Deep Learning Framework 3D-AI, Fraunhofer IPM itself provides various tools - e.g. for annotating 2D and 3D data. After creating a training data set in cooperation with our customers, the ANN is trained on the concrete structural feature classes. Little by little it learns these classes and recognizes them reliably itself at the end.